How Paragon Built Helm-Only Deployment to Unlock Athena Intelligence's Enterprise Customers

About Athena Intelligence

Athena Intelligence was built by people who already knew what it takes to ship software inside the world’s most security-conscious organizations. The founding team came out of Palantir, SpaceX, IBM, and Scale AI, and CEO Brendon Geils spent years deploying mission-critical systems in the public sector before starting the company.

That background shows up in the product: an AI-native workspace, running on Olympus, built for the analytical work of enterprises in finance, law, defense, audit, and consulting. It is SOC 2 Type II certified and HIPAA compliant, and it runs forward deployed inside each customer’s own cloud, where users work alongside AI agents to do end-to-end analysis.

Running inside the customer’s environment is the constraint that shapes everything else. Athena’s customers will not let an outside system reach in and touch their data, so the platform and its integration layer both have to deploy inside the customer’s walls. When Athena went looking for an integration platform that could do that, it came up almost empty.

The data has to stay inside the customer’s cloud

For the enterprises Athena sells to, an integration layer cannot sit outside the firewall and call in. It has to run inside their environment. By Athena’s account, that single requirement ruled out nearly every integration vendor it looked at: most do not deploy on-premises or inside a customer VPC.

Paragon did. It could run the entire integration layer inside the customer’s own cloud, alongside Athena’s software, with no split between what Athena ran in Paragon’s managed cloud and what ran on-prem. For Athena’s engineering lead, Philippe Clesca, that flexibility was the reason Paragon made the shortlist at all.

“Where Paragon really fit into our business model and our paradigm is the ability to go on-premises and actually deploy Paragon along with our software within customers’ environments. Having that in a way where we’re not segmented between our managed cloud and our on-prem VPC deployments was a huge selling point for us.”

A Path that uses Helm alone

Paragon’s standard setup leaned on Terraform, an infrastructure tool that runs a script to create the supporting pieces a deployment needs, like its Postgres database and Redis cache, before the software goes live. Athena’s customers would not allow it. They provision infrastructure with their own internally vetted scripts to stay inside their security policies, and letting an outside installer run against their environment was off the table.

That left the engagement stuck. Without another way in, “quite frankly, yeah, the deal would have been dead.” Clesca said.

The way through was already inside Paragon’s installer. Terraform was only the first step. At the end, it produces a Helm chart, the package Kubernetes uses to run the software, and Athena’s customers had no objection to the chart itself. Their restriction was narrow: no outside script provisioning resources in their environment.

So Paragon built a path that skipped Terraform. The customer supplies their own database and cache, and the Helm chart handles the rest. Nothing external touches the customer’s infrastructure, and they keep control of it without Athena maintaining a separate on-prem build.

Paragon built the missing path

Many vendors would have declined the request. Paragon treated it as a design problem. A small group from account management, platform engineering, and deployment worked the new path alongside Athena’s team: Paragon built and refined the Helm-only deployment, Athena ran it in their environments and reported back what broke.

“With a lot of other vendors, they would have just said, ‘yeah, sorry, we can’t.’ But the fact that you guys were able to move and work with us closely on something like this, it definitely speaks volumes.”

Putting it into production

Paragon deploys as roughly 50 pods, each handling a different job, that all have to come online and talk to each other. Even so, Athena had it running in under two weeks of working through configuration and environment variables. Clesca knew the approach would hold when he deployed it cleanly across three separate test environments and felt confident pointing it at live customer infrastructure. From the moment Athena needed a new route to a deployment proven across multiple environments, the work took a little over a month.

That timeline held because the in-cluster version runs the same as Paragon’s managed cloud. Athena did not fork its codebase or maintain a separate on-prem build. Going live meant changing a few environment variables and pointing the software at the right URLs.

What deploying in place made possible

The payoff was a set of use cases that had been closed to Athena. The clearest is email. Routing a customer’s Outlook or SharePoint data through any outside system, even a proxy, was something Athena’s customers would never approve. Running inside their environment changed that, and connecting to email and file storage in place became possible for the first time. Those connections now anchor much of the work Athena’s agents do.

Clesca said Athena’s platform has grown roughly tenfold since onboarding with Paragon, “maybe 10 times the number of users from that point to where we are as of right now.”

The build-versus-buy call

Athena has a strong engineering team and seriously weighed building the integration layer itself. It decided the cost was too high. The first connector is the easy part. The real expense is everything after it: maintaining a large catalog of connectors, authentication flows, security reviews, and the steady stream of third-party API changes. For a small, fast-moving team, that was the wrong place to spend engineering time.

Clesca’s advice to other engineering leaders facing the same decision is blunt.

“Don’t build it yourself, get Paragon. It’s already baked in and easy to work with. Why would you build it yourself?”

Building AI for regulated enterprises? Speak with our team to see how Paragon’s on-prem and VPC deployment handles the integration layer so your team can stay focused on the work.